12 stories tagged by Digg
AI
Anthropic reports GLM-5.3 built end-to-end exploits in 50 of 410 offline test attempts, versus 56 for Mythos Preview.
AI
Baseten’s announcement covers Codex and the Responses API, using existing OpenAI commitments. Nathan Lambert sees a favorable signal for open models.
AI
One post says it’s a hard time for people in AI who want to raise issues, criticize when warranted, update their beliefs and think in public.
AI
Someone who says he worked on GPT-OSS rejects the idea that OpenAI would use safety as an excuse. The person he challenged says multiple former OpenAI workers told him the disputed claim, and argues that safety and Kimi can both be part of the story.
AI
Discusses GLM-5.3 performance and Chinese labs matching frontier models.
AI
Allen Institute researcher shares lectures with open slides to accompany his RLHF book.
AI
Debate focuses on whether labs can control interfaces as tokens commoditize.
AI
Discussion covers research ideas transitioning into frontier models along with data and sequencing issues in DPO.
AI
License restricts commercial use above revenue thresholds while allowing free non-commercial access.
